Superplane App Builder

SuperPlane App Builder

Translate workflow requirements into SuperPlane apps and their canvas YAML.

Quick Reference

Order of Operations

Always follow this sequence. The CLI is the primary path — it gives exact names, IDs, and schemas that documentation cannot reliably substitute.

1. Verify CLI and Connect

command -v superplane

If this does not print a path: stop. Tell the user to install the CLI from https://docs.superplane.com/installation/cli and re-run the task. Do not attempt to install it on their behalf. Do not silently fall back to doc-based design.

Then verify the current session:

superplane whoami

If whoami fails because of authentication, DNS, timeout, or connection issues, the CLI is installed but the session is not usable yet. Ask the user to connect, fix the context, or allow network access as needed. Do not continue without a working CLI session.

If not yet connected:

superplane connect <URL> <TOKEN>
superplane whoami

If connection details are not available, stop and ask the user to connect/provide the required URL and token. Do not continue without a working CLI session.

1b. Apply Changes via Staging or Direct Commit

App versions are immutable commits. Stage edits before committing, or pass --message to commit directly.

superplane apps staging update --file canvas.yaml
superplane apps staging commit --message "Update canvas"

# or commit in one step:
superplane apps canvas update -f canvas.yaml --message "Update canvas"

2. Understand the Workflow

Before running discovery commands, identify what the workflow needs:

• What starts it? → trigger (schedule, webhook, GitHub push, manual)
• What steps happen? → each step is an action node
• Any decisions? → If or Filter actions for branching
• Any waits? → Approval, Time Gate, Wait actions
• Which external systems? → each maps to a provider (e.g., GitHub, Slack, Daytona)

Collect the list of required providers from this analysis — you will check them in the next step.

3. Discover and Verify Integrations

Run superplane integrations list to get all connected integrations in the org. Compare against the required providers from step 2.

If any required provider is missing: stop and tell the user before writing any YAML. Example:

This canvas needs GitHub and Daytona integrations. Your org has GitHub connected but Daytona is not connected. Please connect it in the SuperPlane UI (Settings → Integrations) before proceeding.

Do not generate YAML that references providers the org has not connected — it will fail with "integration is required" on every affected node.

Once all providers are confirmed connected, discover exact names and schemas:

superplane integrations list                          # connected instances → real integration IDs
superplane index triggers --from <provider>           # exact trigger names
superplane index actions --from <provider>         # exact action names

Inspect required config fields, output channels, and payload shape:

superplane index triggers --name github.onPush --output json
superplane index actions --name semaphore.runWorkflow --output json

List runtime options for integration-resource fields:

superplane integrations list-resources --id <id> --type <type>

Hard gate before writing/applying YAML: for every integration-resource field value you set (for example: repository, snapshot, sandbox, project), verify the exact value exists in list-resources output for that integration. If it does not exist, stop and ask the user which valid value to use.

Schema precedence rule: if provider reference examples conflict with CLI schema output (superplane index ... --output json) or current list-resources values, follow CLI output. References are helper material; CLI is source of truth.

4. Select Actions and Wire the Graph

Use the exact trigger and action names from step 3 — not guesses from documentation.

• If the trigger supports built-in filtering (content filter, action filter, ref filter), configure it at the trigger level. Only add a separate Filter or If node when you need logic the trigger's native config cannot express.
• Every action needs at least one incoming edge
• Triggers have no incoming edges
• Use named channels for branching (If → true/false, Approval → approved/rejected)
• Filter only emits to default when the expression is true; false events stop silently
• Use Merge to fan-in parallel branches

See Components & Triggers Reference for the full list.

5. Position Nodes

Every node needs a position: { x, y }. Nodes are 515px wide × 215px tall — use these spacing rules to prevent overlap:

Start the first node (trigger) at { x: 120, y: 100 }.

Linear pipeline — same y, increment x:

Trigger: { x: 120, y: 100 }  →  Step A: { x: 720, y: 100 }  →  Step B: { x: 1320, y: 100 }

Branching — branches share the same x column, spread on y. Center the source node vertically relative to its branches:

                                ┌─ Branch A: { x: 1320, y: 100 }
Source: { x: 720, y: 250 }  ───┤
                                └─ Branch B: { x: 1320, y: 400 }

Fan-in (Merge) — next x column after branches, y centered between them:

Branch A: { x: 1320, y: 100 } ──┐
                                 ├── Merge: { x: 1920, y: 250 }
Branch B: { x: 1320, y: 400 } ──┘

For 3+ branches, keep adding 300 to y for each branch and center the source/merge accordingly.

6. Configure Expressions

STOP before writing any expression that references payload fields you have not confirmed. Do not guess field paths from trigger or action names.

Envelope

Every node output is wrapped in an envelope: { data: {...}, timestamp, type }. All three access patterns return this envelope, so you always need .data. to reach the actual payload:

Common mistake: writing $['Create Sandbox'].id instead of $['Create Sandbox'].data.id. Always include .data..

Important — example payloads already represent the envelope. The example payload printed by superplane index triggers --name <trigger> (and shown in this skill's reference files) is the full envelope: the outermost { data: {...}, timestamp, type }. root().data returns that top-level data object — do not add an extra .data on top of it. To build a correct expression, count the data keys directly from the example JSON: one .data per nested data key you need to traverse, no more.

Nested data keys. Some webhooks (notably Sentry) carry their own data field inside the payload body. The correct path therefore has two .data segments — root().data.data.issue.title — not three. root().data already unwraps the envelope; the second .data walks into the webhook's own data field. If you find yourself writing root().data.data.data.…, you are double-counting the envelope.

Use double curly braces {{ }} for expressions in configuration fields:

{{ $['GitHub onPush'].data.ref }}

How to confirm payload fields

Check these sources in order — use the first one available:

1. Existing executions — inspect real payloads from prior runs (most reliable):

   superplane executions list --app-id <id> --node-id <nid> -o yaml
   

2. Provider reference files in this skill — check the references/ directory for the provider you are using. These contain payload examples and known gotchas.

3. SuperPlane docs — fetch the provider's action page from the LLM-friendly docs:

   • Compact index: https://docs.superplane.com/llms.txt
   • Full content: https://docs.superplane.com/llms-full.txt

After the first real execution, always go back to source 1 to verify and correct expressions. The trigger name does not map 1:1 to payload structure — always check the provider reference file or docs for the actual webhook event a trigger maps to.

6b. Command Node Best Practices

When an action executes shell commands (e.g., daytona.executeCommand, ssh):

• Use the action's native workingDirectory / envVars config instead of inline cd or export in the shell string. This reduces quoting complexity and failure surface.
• Redirect verbose output to a file and emit a concise status marker to stdout (e.g., STEP_OK / STEP_FAILED). Large or binary stdout can cause node processing issues.
• Check the provider reference file (references/ directory) for the shell execution model, hardened command templates, and known failure patterns specific to that integration.
• For long multi-step scripts, prefer YAML block scalar (command: |-) over folded single-line strings to avoid whitespace/newline parse artifacts in bash -lc.
• Before shipping, run one manual trigger and inspect node outputs in execution YAML to confirm expected channel routing (success vs failed) matches your edge wiring.

7. Apply

Generate a starter YAML if starting from scratch:

superplane apps canvas init --output-file canvas.yaml
# or start from a template:
superplane apps canvas init --template health-check-monitor --output-file canvas.yaml

Then create an app from the file or update an existing app canvas:

superplane apps create --canvas-file canvas.yaml
# or update an existing canvas:
superplane apps staging update --file canvas.yaml
superplane apps staging commit --message "Update canvas"
# or commit directly:
superplane apps canvas update -f canvas.yaml --message "Update canvas"

When creating a new app from canvas YAML, apps create --canvas-file already applies the graph in the file:

superplane apps create --canvas-file canvas.yaml

Workflow rules:

• superplane apps create --canvas-file canvas.yaml accepts the resource-style Canvas YAML from the spec (apiVersion, kind, metadata, spec).
• On superplane apps create, canvas layout flags are prefixed with canvas-: --canvas-auto-layout, --canvas-auto-layout-scope, and repeated --canvas-auto-layout-node.
• Only run superplane apps canvas update ... after create when you are intentionally applying additional changes, such as a later file that includes metadata.id, or explicit auto-layout flags different from the defaults used by create.
• Use apps staging update + apps staging commit for iterative edits, or canvas update / console set with --message to commit directly.

Then verify:

superplane apps canvas get <name>

Check for errorMessage or warningMessage on any node.

8. Definition of Done (Canvas Creation)

Before calling the canvas "ready", confirm all of the following:

• Integration IDs resolved from superplane integrations list
• Every integration-resource value verified via superplane integrations list-resources
• Canvas created from the intended YAML payload
• superplane apps canvas get <name> -o yaml shows empty errorMessage and warningMessage on all nodes
• At least one real trigger run checked, including channel-level outputs from critical branching nodes

Common Patterns

Linear: Trigger → A → B → C

nodes:
  - { id: trigger, ..., position: { x: 120, y: 100 } }
  - { id: a, ..., position: { x: 720, y: 100 } }
  - { id: b, ..., position: { x: 1320, y: 100 } }
  - { id: c, ..., position: { x: 1920, y: 100 } }
edges:
  - { sourceId: trigger, targetId: a, channel: default }
  - { sourceId: a, targetId: b, channel: default }
  - { sourceId: b, targetId: c, channel: default }

Branch: If → true / false

nodes:
  - { id: trigger, ..., position: { x: 120, y: 250 } }
  - { id: check, ..., action: { name: if }, position: { x: 720, y: 250 } }
  - { id: on-true, ..., position: { x: 1320, y: 100 } }
  - { id: on-false, ..., position: { x: 1320, y: 400 } }
edges:
  - { sourceId: trigger, targetId: check, channel: default }
  - { sourceId: check, targetId: on-true, channel: true }
  - { sourceId: check, targetId: on-false, channel: false }

Gate: Filter (pass or stop)

Filter only emits to default when true. False events stop — no edge needed.

nodes:
  - { id: trigger, ..., position: { x: 120, y: 100 } }
  - { id: filter, ..., action: { name: filter }, position: { x: 720, y: 100 } }
  - { id: next-step, ..., position: { x: 1320, y: 100 } }
edges:
  - { sourceId: trigger, targetId: filter, channel: default }
  - { sourceId: filter, targetId: next-step, channel: default }

Fan-out / Fan-in

nodes:
  - { id: trigger, ..., position: { x: 120, y: 250 } }
  - { id: a, ..., position: { x: 720, y: 100 } }
  - { id: b, ..., position: { x: 720, y: 400 } }
  - { id: merge, ..., position: { x: 1320, y: 250 } }
  - { id: final, ..., position: { x: 1920, y: 250 } }
edges:
  - { sourceId: trigger, targetId: a, channel: default }
  - { sourceId: trigger, targetId: b, channel: default }
  - { sourceId: a, targetId: merge, channel: default }
  - { sourceId: b, targetId: merge, channel: default }
  - { sourceId: merge, targetId: final, channel: default }

Approval Gate

nodes:
  - { id: ci-done, ..., position: { x: 120, y: 100 } }
  - { id: timegate, ..., position: { x: 720, y: 100 } }
  - { id: approval, ..., position: { x: 1320, y: 100 } }
  - { id: deploy, ..., position: { x: 1920, y: 100 } }
edges:
  - { sourceId: ci-done, targetId: timegate, channel: default }
  - { sourceId: timegate, targetId: approval, channel: default }
  - { sourceId: approval, targetId: deploy, channel: approved }

When to Use Other Skills

Documentation

For agents that can fetch URLs, the full SuperPlane docs are available in LLM-friendly format:

• Compact index: https://docs.superplane.com/llms.txt
• Full content: https://docs.superplane.com/llms-full.txt

References

• Actions & Triggers — Built-in actions and trigger types
• GitHub — Triggers, actions, payload examples, gotchas
• Daytona — Actions, payload examples, gotchas
• Sentry — Triggers, payload examples, nested data.data gotcha